Custom stack, not SaaS. An AI agent for businesses trained on your own content: chatbots with AI over your catalog and documentation, assistants for bookings or leads, and n8n automations that offload your team from repetitive daily tasks. Claude API, vector embeddings and RAG. Documented stack, no black box, no LinkedIn demos.
Custom chatbot + RAG + n8n + deployment | 2-6 weeks | from €1,500
LinkedIn has sold you AI as the solution to everything. Gorgeous demos, €6,000 proposals that say "GPT" when you ask about the stack. And meanwhile, your team keeps answering the same emails 50 times a day.
Agencies sell "AI solutions" without showing any code. Ask about the stack and they answer "GPT". With no verifiable stack there's no AI of your own: there's margin.
50phrases
What a SaaS chatbot knows before it drops the "let me pass you to a human". Your catalogue, your FAQs and your manuals stay out.
70%
Repetitive tasks
that can be automated in an SMB
You have a catalogue, history, FAQs and internal manuals. Gold for an agent with RAG. Nobody has shown you how it connects.
8 in 10 SMBs with untapped potential
Claude API or OpenAI depending on the case, vector embeddings on your own documents, n8n for the workflows and RAG (Retrieval-Augmented Generation) so the agent answers from your knowledge base, not with made-up answers. Integrated into your website or deployed as a widget. No template SaaS, no lock-in.
Ingestion of your documents (PDFs, URLs, spreadsheets, internal FAQs) into a vector base. When a customer asks, the agent finds the relevant chunk and answers with that specific content. Personality and tone adapted to your brand. Escalates to a human with full context when needed.
An agent that collects lead data, qualifies intent, books a reservation or call, routes to the right channel (WhatsApp, email, Cal.com) and leaves the context in your CRM. Adapted to the funnel: a restaurant takes bookings, B2B qualifies leads, an academy schedules admission calls. We design the logic together in the brief.
If your customer searches "something comfy for running in winter" instead of "thermal running shoe", the traditional search returns zero results. With vector embeddings the system understands intent and returns relevant products. Applies to e-commerce, technical repositories, manuals, legal bases.
Self-hosted n8n with Claude nodes and integrated tools. Workflows that classify emails, draft initial replies in your tone, translate descriptions into 3 languages, score leads, send alerts when condition X is met. The repetitive stuff stops costing hours and stops failing.
You don't need a new website. The agent deploys as a JS widget embedded in your current site (WordPress, custom code, Shopify) or as a REST API your team consumes from wherever they want (CRM, WhatsApp Business, internal Slack). If the site is very old, I warn you about the limitations.
Are you an agency and want to offer white-label AI agents? See Partner agencies. Consultative model, no public pricing.
A verifiable architecture,
not a SaaS black box.
2 to 6 weeks
from technical brief to deployment
"Finally an agent that answers from our catalogue. The team stopped fielding the same 50 questions a day and focuses on closing."
Average across implementations with a well-curated knowledge base. The rest escalate to a human with full history context.
Price by scope.
No surprise invoices.
Price depends on scope: data volume to index, agent complexity, integrations with your systems and deployment channels. I send a closed range after the 48 h technical brief. LLM API and vector base costs are separate, tracked in your own dashboard so you see exactly what each conversation consumes. No opaque invoices, no hidden usage pricing.
AI implementation
The questions I get most before implementing an AI agent.
Closed project from €1,500 up to €8,000 by scope (basic FAQ chatbot vs complex assistant with CRM integrations). Setup + monthly retainer from €400/month. Price depends on three factors: data volume to index, integration complexity with existing systems and deployment channels (web, WhatsApp, CRM). LLM API costs (Claude/OpenAI) and vector base are separate and you pay them directly to the provider, tracked in your dashboard so you see exactly what each conversation consumes. Technical brief in 48 h after initial call, always free.
A template chatbot answers with what you type into a config box: 50-100 prefab sentences. If the customer asks anything unexpected, it fires "let me transfer you". An AI agent with vector embeddings indexes all your real documentation (catalogue, FAQs, manuals, history) and when a customer asks, the system finds the relevant chunk and answers with that specific information. It can also make decisions (book, schedule, qualify lead), escalate to human with full context, and learn from feedback. The technical difference is called RAG (Retrieval-Augmented Generation).
Between 2 and 6 weeks from signed brief. Simple projects (FAQ + lead capture on web): 2-3 weeks. Medium projects (chatbot + CRM + WhatsApp integration): 3-4 weeks. Complex projects (semantic search + n8n automations + multi-channel): 4-6 weeks. The timeline holds if your team delivers data and validates responses in agreed iterations. If the knowledge base is scattered or outdated, that extends the timeline and I tell you from the technical brief.
Three things: (1) business documentation in any format (PDFs, your website URLs, spreadsheets, catalogue exports, internal FAQs in Notion or Google Docs), (2) access to systems the agent must consult or update (CRM, WhatsApp Business, web form, booking calendar), (3) a reference person on your team to validate agent responses during 2-3 training iterations. If data is scattered or outdated, I help you structure it as part of the project.
No. The agent deploys as embedded JS widget on your existing site (WordPress, custom code, Shopify, Webflow), or as REST API your team consumes from anywhere (CRM, WhatsApp, Slack, internal panel). If the site is very old or has serious technical issues, I warn you from the brief. If we're building it in parallel with Conversion Web, the widget ships built-in without extra integration work.
Yes. The vector knowledge base updates with new documents without retraining from scratch: you add a PDF, a page or an export and it's indexed in minutes. Integrations with new systems (a new CRM, another messaging channel, an extra language) are added modularly. The monthly retainer model covers these periodic updates. If volume grows much (10x conversations/month), we review architecture (dedicated vector base, caching) without replacing the system.
Yes, and being transparent here is part of the deal. (1) It doesn't replace human judgement on sensitive decisions (legal, medical, serious financial): it has to escalate to a human, always. (2) It can hallucinate if the knowledge base has inconsistencies or contradictions, which is why curating the data is part of the process. (3) It needs maintenance when the catalogue, prices or processes change, that's what the monthly retainer is for. (4) It won't give you magical month-1 ROI: like SEO, it multiplies what already works, it doesn't replace it. If someone promises that, they're selling LinkedIn demos.
20 min and no commitment. You tell me what the agent has to solve for your business (support, bookings, scoring, search), what data you have on hand and what channels you need. In 48 h I send you a written technical brief with proposed architecture and closed price range. If it doesn't fit, I point you elsewhere without sugarcoating.
Book 20-min call20-min technical call.
You and me, no stackless salespeople. You tell me the business problem to solve and what data you have.
What happens next.
Written technical brief in 48 h with scope, architecture, price range and timeline. You decide calmly.
